Style Vectors for Steering Generative Large Language Model
Published in EACL '24: Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: Findings, 2024
Recommended citation: Konen, K., Jentzsch, S., Diallo, D., Schütt, P., Bensch, O., El Baff, R., Opitz, D. & Hecking, T. (2024). Style Vectors for Steering Generative Large Language Model. arXiv e-prints, arXiv-2402. https://arxiv.org/abs/2402.01618
Introduction of a new approach to efficiently derive style vectors from LLM activation layers. We provide empirical proof of the effectiveness and efficiency of activation engineering steering approaches, and compare them to training-based style steering.
Recommended citation: Konen, K., Jentzsch, S., Diallo, D., Schütt, P., Bensch, O., El Baff, R., Opitz, D. & Hecking, T. (2024). Style Vectors for Steering Generative Large Language Model. arXiv e-prints, arXiv-2402.